Optimized Infusion Scheduling: Your EHR Can’t Do This Webinar Transcript
MOHAN: Welcome, everybody, to– this is the third webinar in the four part series. The way we’ve laid out these webinars, each one progressively focuses on a different topic. So webinar one focused on why the EHR is not built to optimize scheduling, and why the mathematics of it are harder than one might think. Webinar two focused on, how do you actually optimize a schedule?
Today’s webinar will focus on how you shape the future performance of your infusion center. And the next one will focus on the diagnostics. Just in the interest of the folks who did not attend the first two, I’ll quickly do a refresher. So within the first 10 or 15 minutes we’ll cover all of the prior topics. Those topics are available in greater detail in the recorded webinar, so feel free to go download those and watch them in greater detail.
- So today we will start with a quick introduction of who we are, then spend five minutes in total talking about why EHRs are not built to do this for you, talk about infusion optimization in five minutes, and then spend the bulk of our time talking about planning.
- So who is LeanTaaS and what do we do? We are a Silicon Valley software company. And our focus is on using mathematics and software to unlock capacity in hospitals. And the magic of unlocking the capacity is that many good things follow. Patient access improves, meaning it’s a shorter lead time to future appointments. Wait time for patients go down, which is the biggest source of patient dissatisfaction, and so having the capacity to require patients to wait less is a very good thing. Both operating costs and capital costs improve, because the ability to use the staff and the assets more productively just results in an overall lower cost envelope. And finally, to the extent that the capacity unlocking allows you to see more patients. There is a revenue uplift from that as well.
The two products that we have commercially out there are Infusion and. Operating Rooms. And we are in the middle of working with partners on building the next set of products. So for example, we are working with 20 oncologists each at Memorial Sloan Kettering in New York and M.D. Anderson in Houston, to build our oncology template optimization for the providers, so that the wait times and clinics can get reduced. Similarly, we’ve got parallel initiatives on imaging inpatient beds and labs. In terms of who we work with, these are the leading health systems across the country, and they span the gamut from academic medical centers like Stanford and UCSF, and Duke and Emory, and UPenn, to iconic institutions like M.D. Anderson, Sloan Kettering, and Johns. Hopkins, to regional hospitals and cancer centers as well.
The fact that infusion scheduling is a pinpoint felt by many, many cancer centers is evident in the fact that from 0 or 1 infusion center in early 2015, we’re now running 113 infusion centers with nearly 3,000 shares, based on the optimization algorithms. So that’s us in a nutshell. To give you a quick refresher on why is it that the EHR is not built to optimize, there are three big reasons for this. The first is, as you look around any health system, there is a tendency to rely on a grid-based schedule, which is in point one, where the assets are laid across the top and times of day down the left. The assets could be chairs, rooms, providers, imaging machines, whatever the asset is.
And so when John Smith needs an 8:00 to 9:00 appointment, someone somewhere colors in 8:00 to 9:00, and puts down John Smith’s name or MRN number, this is on spreadsheets, on whiteboards, on snap boards in the EHR, et cetera. This just does not work. The reason is grid-based scheduling works if you’re scheduling something that is deterministic, meaning the start and the end time is known at the time of making the appointment, as is true for tennis courts and spa treatments. Medical appointments and infusion treatments are stochastic, meaning they are random and highly variable. The start time and the end time doesn’t work out as you thought, and therefore using a deterministic scheduling framework, like a grid, to schedule a stochastic thing like an infusion appointment, is just flat out mathematically wrong.
This is why the grid looks great on the previous night and the day never works out as planned. The second reason, the mathematics of EHR is simply not robust enough. Every EHR, and therefore, every health system, follows a first come, first schedule sort of a discipline in booking appointments. Meaning, if you call and ask for an appointment seven months into the future, you typically will be told the calendar is open, pick a spot, any spot. Here are our hours of operation. That looks nice and patient centric. It’s wrong because nothing mathematically balanced the load. If you’re going to balance the load, then continuously throughout the day you have to balance the number of people just getting started, with the number of people just leaving, and the number of people in between. That’s the way you create a balance profile, first come, first schedule just does not allow that to happen. And the third thing that’s mathematically, completely inadequate in EHRs, is this notion of a connecting appointment. Almost everything in a health system is a connected series of appointments, labs followed by clinics, followed by infusion, followed by radiation oncology.
When connected schedules are built, they have to first off ensure that the earlier segments operate on time, just like your first flight has to be on time in order for you to make your connecting flight. That doesn’t often happen. And the second thing is, these connected schedules are built one person at a time. So John Smith is given the 8 o’clock appointment at the doc and the 9 o’clock appointment at the infusion, and then along comes Jane Doe and the same thing happens. Connecting schedules need to be built optimizing the system as a whole, not one person at a time. So when you think about how connecting flights are put together, they don’t sort out the departure time of the second flight based on one passenger at a time. They optimize the whole system. So for these reasons, the mathematics is just simply not adequate.
Now, infusion at the surface looks like it should be simple. It’s a person sitting in a chair, getting infused for some number of hours, from one to nine hours. And so on the surface it says, how hard can it possibly be? Well, it turns out it’s actually very, very hard. Even if you took five types of appointments, one hour, two hour, three to five, six to eight, and nine plus, assume a modest sized infusion center, 20, 25 chairs that sees up to 70 patients a day. With this duration mix, followed by how many possible slots can there be, if you can offer up appointments at 10 minute intervals 7:00, 7:10, 7:20, and so on, you’ll have 64 slots. And if you could seat four patients at a time, that’s 256 possible slots. Mathematically, that’s a number with 105 zeros behind it. That’s the number of permutations and combinations for how you could build that schedule. A number with 105 zeros behind it is staggering. Just to put this in context, the odds of winning the Mega Millions is 1 in 259 million, the odds of winning. Powerball is 1 in 292, if you wanted to win both of those, one after the other, that’s a number with 15 zeros behind it. So you’d have to win both of them, seven times in a row, in order to get the infusion schedule right for just one day. That’s how mathematically overwhelming it is. And so expecting schedulers and nurse managers and other clinic folks to just pick a number, pick a slot on the calendar, and get it even remotely right is just not going to happen.
After looking at all these permutations, what makes it harder is there are real life operational constraints. The volume and mix for each day is unique to a day of week. The nurse availability and workload, which we’ll spend more time on, is unique. The chair availability also depends. And finally, there’s lots of expected and unexpected variability of clinics running late, which you could have predicted, and clinics running late or patients running late that you could not have predicted. And so all of these have to be taken into account. Because of the maps not being sufficient, in most infusion centers, the day plays out like this. In a peaky profile, where patients arrive roughly in the order of their appointment with varying treatment lengths needed, the charge nurse tries to put them in the right part, in the right chair, with the right nurse. And the game of Tetris as it unfolds is a losing hand. The reason this matters is the duration of the peak is only three or four hours. And the duration of a nursing shift is eight or 10. Which means, right off the bat you’re confronted with a bad choice. Should you staff for the peak? In which case you are overstaffed before and after. Or should you staff for the average? In which case, at the peak, right when you need it most, you are understaffed.
But the more chronic problem is that any time a peak approaches capacity a system becomes mathematically unstable. Think of the freeways at rush hour, it’s approaching system capacity, and therefore, every metric goes into the tank, which is what makes it mathematically unstable. A 10 minute drive takes 60 minutes. A fender bender that should take 10 minutes to clear takes two hours to clear. And a fender bender that should delay 10 people delays 10,000 people. Infusion is a series of fender benders waiting to happen. The clinic will run late, the pharmacy will back up, the labs will back up, a nurse will call in sick, a patient will show up late, a patient would react badly. If any of those happen early on in the day or late in the afternoon, it’s fine. If any of them happen in the middle of the day, it’s like the big rig crash at 5:00 PM on a Monday, the whole system will be a mess for many hours. Because of all this, most infusion centers have uniquely the same three problems. Patients wait a long time, particularly in the middle of the day. The chair profile starts out narrow, hits a peak, and comes down, sometimes it exceeds the chair capacity. What that means is all the chairs in an infusion suite are full, and some chairs in the waiting room are occupied as well. And nurses tend to either miss their lunch, or get a granola bar or an apple and eat lunch at 4 o’clock, or centers over-staff. They get float nurses to cover the lunch break, which is essentially throwing nursing labor at the problem, not really solving the underlying problem.
So that covers the first two topics. In four or five minutes let me just cover what an optimized infusion schedule looks like. In order to optimize it, what you have to do is take this profile and make a dash profile. And in order to do that, there’s a bunch of hard math problems that need to be solved. First, with very high accuracy, predict the volume for a day of week. How many on a Monday? How many on a Tuesday? Volume in aggregate is not enough. You have to predict the mix very accurately. How many one hour treatments? How many two hour treatments? How many three hour treatments? Having done that, you then need to figure out how much adjustment you have to make. So by looking historically at the last 1,000 times, you thought something would be three hours, and then seeing how accurate you were, will help you understand how much you need to adjust it. And then having adjusted it, you have to play the supercomputer version of Tetris, which is how do I arrange all of those blocks such that I generate the flattest profile possible?
The reason this becomes magical is it unlocks capacity. Unlocking capacity during peak hours is like magically finding three extra lanes on the freeway that you weren’t expecting. It gives patient choice. Meaning, if I needed a three hour treatment there are many three hour trains leaving the station. It flattens the workload for nurses which allows them to be more productive throughout the day. And it fits the nursing schedule. Nurses who show up first, leave first. Nurses who show up last, leave last.
So this is how the optimization needs to work. So what needs to happen, after all of this, is to create a template that’s unique for every center, for every day, for every hour of the day. That’s how precise and fine-grained the template needs to be. And so the way you do that is think of a simplified grid where across the top we’re laying out the duration buckets, one hour, two hour, three to five, six to eight, and nine plus. We are then saying 10 minute starts, so 7:00 AM, 7:10, these are the number of patients to start. The capacity of the infusion center limits how many simultaneous starts it can do. This center said between 8:00 and 8:30 they could see three or four patients at a time. Once it’s past 9 o’clock they can only see two at a time. So no row after 9 o’clock has more than two starts. The duration spans are broad enough to allow for variability. And the magic number is for every duration bucket, for every start time, how many patients of that type should you start.
And so now, the dialogue with the patient becomes slightly more nuanced. Rather than saying, Mrs. Jones, when would you like to come? And then just accepting that time. The conversation goes, I see you need a three to five hour treatment. I could offer you 8 o’clock, 9:20, 9:40, 10:00, 10:20 or 10:40, would any of those work? And so you don’t have to get 100% of them right, but if you get 80, 85% of people into the right buckets, it’s pre-engineered to give you a flat profile. Right? The reason this unlocks capacity is if you look at the before and after of the way the Tetris blocks loaded up, all of this white space here is lost productivity, it’s either under-utilization of a chair, or under-utilization of a nurse, or both. On the right hand side, because you’ve mathematically tightened the Tetris blocks taking into account variability, you unlock productivity.
The productivity can be measured as the ratio of patient hours to nursing hours. So patient hours is the number of patients times the average duration. Nursing hours is the number of nursing FTE hours on the floor. If this ratio gets better, and we can make it better by 20% or 25%, you now have choices. You could choose to monetize that productivity by absorbing more volume, which means you either see more patients in the day for the same number of nursing staff, or you absorb future growth at either the same staff level or adding less than you otherwise might have. That’s the way you monetize by absorbing growth, or you could monetize by reducing cost. Meaning, I no longer need to stay open until 9 o’clock, I could close at 6:00, or I could no longer need to keep a satellite center open because my main center can take on more. So you manage to reduce cost, or you could just improve the patient experience, meaning they wait less and they get appointments set at a shorter lead time. So these are choices on monetization.
But the key thing to monetize is it comes after you create the productivity in the first place. OK. So that was the quick refresher of all of the topics we’ve covered in prior webinars. So let’s focus now on the central theme of today’s webinar, which is how can you plan well enough in advance to run the infusion center. Infusion center operations are complicated, it’s highly variable, it’s highly dependent, day for day, hour for hour, patient for patient. And so any advanced intelligence you can get will let you run the infusion center much, much better. So at the very least, you need to know, with a fair degree of precision, how is today going to unfold. And by a fair degree of precision, I mean at 10 minute intervals. So what you’re seeing here is if you run the optimization. The green line represents what would be optimal. That’s if you played the. Tetris game right, that’s the profile you should get. The bars represent 10 minute windows. Gray means you’re right on the optimal frontier. Yellow means you’re running a little bit light, you’re under the optimal frontier. If you’d gone above the optimal it would have been orange.
And if you’d gone above chair capacity it would have been red. That just means all your chairs are full for some period of time. To get this right now, you can plan your day. If you’ve got this at 6:00 in the morning, saying, this is how my day is going to unfold. You immediately know that between 10:30 and 12:20 you’re going to be running right on par. It’s probably better not to do stand-up meetings or have people take extended breaks at that time, kind of run it with an eye on the operation.
If you needed add-ons, you know the two windows between 7:40 and 10 o’clock, and again between 1:45 and 1:40, and the rest of the day you could see your add-ons. This is how today is going to unfold. Equally important is knowing how the next 30 days are going to unfold. So imagine if you could get the weather map for the next 30 days showing you how the systems are forming. So if you can already tell that 10 days from now, the middle of the second row, I’m going to have a little bit of heat where I can see some orange shaping up. You can drill down into that, and start to say, now I understand what’s going to happen. My ramp is slower than. I would have liked, but it’s going to lead to congestion for a 90 minute window in the middle of the day.
Maybe I should either get a few nurses to start a bit later, maybe I can move a patient to start earlier, even though that’s not a preferred thing, or maybe I can just tell my schedulers to not schedule any more because I’m already running pretty late. So this is the sort of stuff that if you had advance warning you can run the infusion center much, much better, versus a reactive mode of having the staff showing up in the morning, knowing that the tsunami is going to hit sometime in the middle of the day, but not knowing when or why or for how long. That makes the difference between a reactive operation and a proactive planned operation. So what does it take to build a planned operation?
There are two very different, but very difficult problems to solve. The first is a supply demand matching problem. What you’ve got to do is, it’s not sufficient to match the demand across the whole day saying, yep, I can deal with 80 patients in the day. Well, it matters whether those 80 patients came eight per hour for 10 hours, or 20 per hour for a four hour window in the middle of the day. So you’ve got to understand the arrival pattern, volume, mix, timing, and try and match it with your supply capacity, staff, equipment, chairs, et cetera, within a very tight window, ideally 10 or 15 minutes, maybe up to 30 minutes, despite the fact that the variability would be very, very high. So that’s problem number one to be solved. Problem number two to be solved, is how do you think about connected services? When you think about optimizing connected services, there are many connected appointments, labs and clinics, and procedures. They have to be spaced far enough to be able to execute it on time every day, but close enough that it’s convenient for patients. If you schedule a connecting flight through O’Hare every day with a 15 minute gap, you know they’ll miss the connecting flight. If you schedule it with a six hour gap, yes, they’ll make the connecting flight, but it’s not efficient or convenient for the passengers. So this is the mathematical balance you have to strike, close enough to be executable, but not so far that it’s inconvenient.
So let’s take each problem in turn and describe how it needs to be solved. To match supply and demand, here’s what needs to happen. You have to analyze the profiles, forecast the volume and mix, and then shape it. Shape the demand profile by sequencing the right number of appointments, off the right duration, at each time slot on the day. That’s kind of what we’ve spent our time up until now talking about, which is how do you build that magic template grid. The other side of the coin is sorting out the supply, nurses and chair availability. And what you’ve got to do is figure out the staff ramp-up schedule, and how you’re going to allocate chairs, right? So if you do those two right, then what comes out of it using the mathematics are two sophisticated models. One is a forecasting model on the volume side, and the second is resource roles and constraints on the supply side. From those, you get this optimization to synchronize demand and supply in very tight windows. So that’s kind of how you work the first part of it, right? Now, what are the supply– so we’ve talked about the demand side. Let’s focus on the supply side.
When you’re going to figure out nursing capacity, which is supply constraint 1, the nurse capacity, you start out by saying, what is the typical staffing profile that you’ve got? How many shifts? How many RNs on each shift? Are the nurses specialized for specific things, like port flushes or port draws? Et cetera. Do MAs and LPNs take patients back? What do charge nurses do? So you have to understand what is the resource pool that you have that you’re working with? And then second, overlay [? on ?] your clinical model. You should not have to change anyone’s clinical model at all, because if people say– if a particular center says, I like my nurse to be one on one with the patient for the first 45 minutes of the treatment. Well, that’s the way they practice and they have to be allowed to do that. So what you’ve got to do is say, how many patients can a single nurse see at one time in mid-flight of the treatment? How much one on one time do you want at the beginning, in the middle, and at the end? How do you think about lunch? Do you have rules, like no more than two nurses at a time taking lunch? Do you spread the lunch break out between 11:00 and 2:00? What exactly do you do? From all of this, you get the nursing parameters.
And what we’ve done on the demand side is teed it up, so that you’re starting the right appointments at the right time to match this nursing capacity. You cannot start a patient unless the nurse is available. Having a chair available is insufficient to start a treatment. You need both the chair and the nurse. And therefore, you have to orchestrate how you started such that both the chair and the nurse are available. If you do that right you will get the flat profile we are talking about. And the speed of the ramp-up is entirely dependent on your nurse availability at the beginning of the day. During the flat portion it’s entirely dependent on how many simultaneous patients can a single nurse support.
Those two constraints determine how high you can go. So you may have 40 chairs, but if you have only two nurses, and say that no nurse can watch more than four patients at a time, you will never get– the flat part of the curve will never be higher than eight. So it’s independent of how many chairs you’ve got. The chairs are one constraint, the nurses are another constraint. So having done that, now you’ve got to think about how nursing shifts need to be made, and how do you assign patients to nurses. Most conventional systems end up just setting up nurse shifts with simple averaging and Excel like things, and say, all right, let’s have two shifts one at 8 o’clock and one starting at 8:30. Let’s get some number of nurses at 8:00 and some others at 8:30. So this is what the profile looks like. That looks fine when you’re just thinking of it from an HR and a labor planning perspective. When you overlay it with how your demand builds up, you’ll see you are very under-utilized in early morning and you’re potentially over-utilized at points in time, where the nurses simply do not have enough capacity to deal with their workload. That creates the rush hour and domino effect problems, which then linger on downstream and for many hours after that.
Many people try and overlay an acuity model on it. And acuity models are great. The problem with most acuity models is they tend to be very aggregate, and they tend to be very inexact. And so what happens is, people assign an acuity score to one to five, and then say, all right, we want each nurse to have 20 acuity points in the day. That sounds like a good and fair thing to do, but it makes a big difference whether, as a nurse, I met my 20 acuity points [? to ?] 20 patients of acuity point, one each, or whether I met it through four patients of acuity points, five each. And so those are wildly different.
The other part that’s wildly different, is it depends on when my points are getting accumulated throughout the day. If all 20 of my acuity points happened between 11:00 and 2:00, I had a very different experience from a colleague nurse who’s 20 acuity points are spread between 8:00 and 6:00. And so using an aggregate, blunt instrument, like an acuity, often feels good, but doesn’t actually do the job. If you’re going to use the acuity points, they need to be used in a very precise, very fine-grained manner. So what needs to happen is, when you lay out the loading profile, you’ve got to try and set up your shift schedule to as closely mirror as possible, both the ramp-up and the ramp-down phases. This may require you to set up more discrete start times, like nurses who start at 8:00, 8:30, 9:00, 09:30 and 10. So you may have more shift starts, which seems to create complexity from an HR standpoint, but it’s important to balance your workload. And then, you need to be able to match which nurse covers which patient in a much more precise way. It’s a holistic workload that depends on acuity. It depends on the number of starts.
It depends on the number of simultaneous patients. You cannot use any one of these metrics to balance out how the nurses get assigned to patients, and how their shifts start. So that’s how you deal with supply constraint 1. If you’re going to deal with supply constraint 2, which is your chairs, what’s happening is the ideal profile is this trapezoidal shape, where ramps up stays flat, and ramps down. You’ll never get a rectangular shape. Because in order to get a rectangular shape, it would have to be like a shotgun start, all 40 infusion chairs are loaded at 7:00 in the morning. There’s no chance that that will happen. It takes one to two and two and a half hours to get the infusion chairs loaded up. And so there will always be a ramp-up and a ramp-down. The goal is to reduce leakage, which means all the white space between this perfect trapezoid shape, and the actual chair utilization, can potentially be improved. So that’s kind of what you’re gunning for, to minimize this leakage or waste.
The second thing you can do is if you want to get more out of your chairs, think about how you speed up your ramp-up. And there are many things you can do with it, one is start earlier, two is have more nurses on deck at the beginning. Most people already have more nurses than they need. The challenge is getting enough patients there in the morning. And so if there’s a way to screen out patients who perhaps don’t need a clinic appointment in advance of the infusion, so the more routine follow up kinds of infusions, try and front load those. So there are various things you can do to ramp-up at the front end. The other thing you can do is sometimes infusion centers ramp-down very gradually over a period of time. If you could make the ramp-down steeper as well, then you’ve got more area to fill in the trapezoid. So think of the more you color inside that trapezoid, the better you are using your second constraint of chairs. And the final thing you can do, at some point, push comes to shove, and you are trying to put 10 pounds into a five pound bag, and there just isn’t enough capacity. So once you recognize that, the first thing you can do, which is a lower cost is to just run more hours, run an extra hour in the evening, run two extra hours in the evening, perhaps run weekends. So you’re basically moving the rectangle to the right, which therefore makes the trapezoid bigger.
And then finally, push comes to shove, you just add more chairs, which makes the height of the rectangle bigger, which means the height of the trapezoid would be bigger as well. So the trick is to understand how you can get the most out of your second supply constraint, which is chairs. Now, in order to analyze it it does require you to do fairly sophisticated simulation math. This is not averaging it out, and saying I get two point five patients per chair, therefore if I add, you know, two more chairs I can see five more patients. That math just breaks down completely, because it tends to be not linear. It tends to be a fairly complicated set of interconnected equations between capacity, volume, and supply. And so what you’ve got to do is build a simulation model that says, what if I added more chairs, what happens? What if I change my hours? What if I change my pharmacy? What if I change my nurse shift? So each of these what if questions you’ve got to be able to run through a simulation engine to then come back and tell you whether it’s a good idea or a bad idea, or a high cost idea or a low cost idea. So that’s kind of how you work with it.
So that’s the first part of the problem, which is solving the demand supply matching problem, treating both nurses and chairs as a mathematical constraint into the optimization equation. The second part is, how do you optimize connected services? This is a totally different problem, nothing to do with the demand supply balance. It’s an equally complicated problem, but it’s a completely different problem. You have to think of it as two sub-parts to it. One is the concept of a linked connected service. When you think of the link connected service it’s the lab, followed by the clinic, followed by infusion. It happens one at a time, and one is done before you can do the second one. You cannot start the second one until the first one is done. This is very similar to planning the route schedule for an airline. So if you are sitting and thinking through, what should the 5,000 flights that Delta does every single day to all of the cities around the world need to do? Then you have to think about point to point, which are hubs, which are spokes? How much connecting time do you need to leave in major gateways, like Atlanta and New York, and Paris for Delta? So you have to think through the entire route schedule, and therefore the appropriate layover durations.
The second problem is a dependent connected service problem. What this means is, all of these pre-things have to happen before this service can occur. So authorization has to happen, and labs have to happen, and the pharmacy has to finish mixing the medications before you can actually execute the infusion. This is very similar to turning around an aircraft at the gate. So when a plane pulls in, lots of things need to happen. It needs to get refueled, it needs to get catered, it need to get cleaned, baggage handling needs to come in. And so all of that has to happen on time in a tight window. So [? shooting ?] to see the patient in a chair, you have to think through, will all these dependent services happen on time, every time, reliably, so that my chair start time is as close to reality as possible? So these are the two different network connected problems. Now, in both of those, you notice I focused a lot on time. Meaning, will the linked appointment happen on time? Will the dependent service complete on time? So why this focus on time? It turns out, if you go to optimize any service process, the intensity of the focus is entirely on turnaround time. And here’s why, when you think about a service, any service, there are only three parameters that matter. How much time did it take to get that service done? What did it cost on a per unit basis? And what was the error rate? And deliberately, you’ll see why. I’m calling it error, and not quality. So these three corners capture, in essence, what the current performance is. And what you’re trying to do is push all three inwards to get to the best performance possible. Lower the unit cost of service, lower the cycle time, and lower the error rate, which is why I wasn’t calling it quality. I just want to go– everything towards the inside.
Now, with this is a framework, let me lay out five assertions. The first is any service process, can be measured on these three metrics, time, cost, and error rate. That’s it. What you call time, could be turn around time, total time, et cetera. Cost could be fixed cost, variable cost, total cost, it doesn’t matter. These are the only three dimensions that matter. The second is the notion that there is a trade-off is completely wrong. Simultaneous improvement across all three is not only possible, is very, very achievable. The trade-off is false [? until ?] you approach perfection. If you’re executing a service at a [? Penneys ?] or a [? Pop, ?] turning around in a millisecond with a 0.0001 error rate, it’s possible you’ve reached the limits, and you cannot make it much better. There are very few services that achieve that goal, and therefore, you can improve all three simultaneously. Now, if you’re going to improve all three, which one should you start with? Most people start with cost. But cost is absolutely the wrong place to start. Service processes are labor intensive processes. If you’re going to try and take the cost out, you’re going to have to take headcount out. And if you take headcount out, what happens is you lose skills and you lose capacity. And when you lose skills, more errors happen. And when you lose capacity, the time takes longer.
So while you may have succeeded in pushing in this cost button of the red buttons, the other two buttons have popped out in the wrong direction. And so it’s a water balloon effect, you push it in one place, it pushes out in the other two. Error reduction is also the wrong place to start. Because an existing process, if you overlay error reduction on it, is only achieved by inspection. And when you try and achieve it by inspection, you put in checkers, and then you put in checkers that check the checkers, who check the checkers. And what happens with that is both the costs go up because you’ve got more checkers involved in the cycle, and the cycle time goes up because it needs to go through more approval phases for anything to get done. So again, it repeats the water balloon effect that you pushed in the error button, but the other two buttons popped up. Time is the only one that when you push it in it forces the other two in.
If you suddenly insisted that the turnaround time has to be half what it used to be, guess what? You no longer have the luxury of making mistakes and redoing it. It had better be right the first time, and therefore pushes you to put into place error prevention mechanisms, and enter detection mechanisms that minimize it. If you’re going to say the cost has to go down– if you’re going to say the time goes in, it forces you to think about a workflow that minimizes hand-offs, because hand-offs mean more people, more people means more cost. So time is the only one when you push the button in, instead of repelling the other two buttons, it pulls the other two buttons in. And that’s why it’s the right place to start. Now, let me make this come to life with an analogy. When you think about manufacturing– what happens is, oftentimes the problem is more invisible than visible. So think about manufacturing. If I had a lot of inventory, then I could have a lot of problems in my process. I could have bad suppliers, I could have production quality problems, my machines could break down all the time. But anytime I got an order to supply 100 widgets, I had a million widgets sitting in my warehouse and I could ship from my inventory. And so what the inventory did, just like the tip of the iceberg, it allowed me to get very sloppy on a bunch of the other metrics and seemingly appear to meet performance objectives even though my underlying manufacturing process was broken. If you took away my inventory, then suddenly when I get an order for 100 widgets, I’m not able to produce it because my suppliers are bad, my compliance is horrible, my production is terrible, and my equipment utilization is bad.
Similarly, in a service process, cycle time is what drives it. So if I have a long cycle time, I can have all kinds of problems below it. So imagine if I took 30 days to process a bank loan. I could get my forms wrong five times. I could get my inspections wrong. I could get the application wrong. Because the amount of work required is only two hours of work, and. I had 30 days to deal with it. So when you force the time to zero, you force a superior performance. So that’s why in both connected services and dependent services you should focus on cycle time. Are you meeting the turnaround time expected for labs or pharmacy? Et cetera.
Let me give you an example. For labs, when you try and say I want to minimize the turnaround time off of a lab blood drop, then the three variables that are involved is, at what rates are our patients arriving into the lab? How many staff do you have to deal with it? Meaning, phlebotomists and nurses et cetera. And how quickly are you processing each patient? These are the only three variables that matter. And there are very tight mathematical equations that connected, right? They’re messy, nasty equations, but they’re very well-defined equations. And when you do this, it lets you balance supply and demand in a 15 minute window, which ties back how connected services play back into the demand and supply balance, which is why you needed to solve both of those problems. At a large academic medical center, we did this for labs. Their issue was the labs were taking an hour to process, and it was regardless of whether they were getting it from the needle or from the port. And so we first analyzed the arrival patterns by hour of day at 15 minute intervals. So we knew how many port draw patients versus regular phlebotomist patients are coming in, took a whole bunch of actions, some of which they had done better. And having done it better, we were then able to give them an optimized schedule of when nurses should start, when phlebotomists should start, give them the training to minimize the turnaround time for processing an order. And as a result of all of those actions, here’s what happened over a fairly short period of eight to 12 weeks, the turnaround time for lab draw went from 50 to 60 minutes, down to 15 to 20 minutes, so 60 plus percent reduction in time, with no drop in quality or cost. In fact, improvements on both of those.
So that’s kind of how all of these pieces fit together. The webinar coming up in a couple of weeks, is on diagnostics. Let me give you a two minute preview of what you can expect to see. In order to analyze the performance, there’s a whole bunch of intelligent diagnostics that need to be done, that start with understanding the volume pattern, understanding the mix. How often is the volume pattern getting close to the edge of the cliff? Because when you get close to the edge of the cliff, is when the wait times go up and so on. How much is the add on and cancellation rate? Which means the inherent variability and instability with schedule. Is it impacting you adversely or is it a modest enough shock to the system that you can absorb? You need to understand that by being able to look at it. You need to be able to tell, are patients arriving early or late, and what’s happening with that. And finally you need to tell whether the guidance on how to steer scheduling is being,. A, followed, and B, executed. And if it’s being followed and the results are still not good, then maybe the template is wrong. If it’s not being followed, and the results aren’t good, then maybe you should follow it first. And once we follow it we can start to see if the template is good or not. And so you need to build a closed loop learning system, which starts with the optimization puts out a guidance for how you should schedule, monitors how you’re actually scheduling against that, connects it to the performance, and then learns from it and tweaks the template.
When you build this learning loop, it’s a bit like a thermostat that knows when the temperature’s above what you set it, as it needs to fire up the air conditioning. And when the temperature’s below what you set it as, it needs to fire up the heating system. And it constantly alternates between the heating and cooling, until you get it stable at the desired temperature. The desired temperature for what you are shooting for an infusion center, is that perfect trapezoid of a smooth ramp-up, flat throughout the day, and a smooth ramp-down. So that’s the content for today. In the last five or 10 minutes, let me focus on incoming questions. The first question is, are the previous webinar slides available? Yes, when you– the same link that you used to register has the previous webinars available for downloading. And so webinars 1 and 2 are available for downloading, which would give you the more detailed views of what I rushed through in the first five or 10 minutes.
And after today, by sometime tomorrow, today’s webinar will be loaded as well. So within 24 hours after the end of the webinar it gets loaded up. OK. Question. Our nurses are unionized, is it still possible to optimize schedules? Yes, very much so. Even in a unionized environment you can set shift schedules and so on. Now, you can’t change them day to day and week to week, you have to plan it well in advance and get approvals and so on. So you can do that. And so what will happen is, if you build up a system of just 10 and 12 hour shifts then you’d want a bunch of eight hour shifts, then that might take you a little bit longer to negotiate with the union to get the right composition of nursing shifts.
But as long as you offer up enough lead time, and give people the choice to work a shift that they want to work, the unionized environment does not necessarily limit to you. New question, what are the challenges around the add-ons to the current schedule? Yes, infusion is a fairly high add-on, and not so high, but suddenly occurs, no-show and cancellation rate. We typically see no-show and cancellation rates of 6% or 7%, and add-on rates roughly that or sometimes a little bit higher. Now, what needs to happen is neither an add-on, nor a cancellation, is precisely predictable. You can predict that x number are likely to, but that’s not helpful, because you don’t know when it’s going to happen. So what you need to do, is for each day of the week, you have to get a very good sense of, is my cancellation envelope higher than my add-on envelope? Meaning, more people are likely to cancel or no-show than are likely to be added on. If so, you’ve got an opportunity. Add-ons should not be dealt with on a first come, first serve basis. Add-ons you need to be very strategic about. If you’ve got to the [? huddled ?] profile that shows when you’ve got pockets of time, then when you get an add-on request, you can confidently steer it to one of those windows of time, knowing you’ll be fine. If you’re trying to steer it into a window that’s crowded, rather than just taking it on, crossing your fingers, and hoping for the best, because that’s what happens every day and it never works out right, is to create some kind of a dynamic wait list. Where you tell the person, Yes, I know you need to be added on sometime in the 12:00 to 3:00 window. I don’t have an exact slot for you yet, give me your cell phone, I’ll text you. Because you know your cancellations will occur, there will be a cancellation or a no-show sometime in that window that happens to match the duration.
The biggest mistake you can make is jam an add-on into an available chair. The add-on could be a five hour opening, and the chair opening based on optimization, is a two hour opening. You’ve now jammed a five hour appointment into a two hour slot that will create a domino effect downstream and mess everything up. So you’re much better off waiting for a five hour cancel to occur, because you know with confidence that a five or six hour cancel will likely occur. And then you steer it. In some ways this is exactly how restaurants deal with it. If you go to a restaurant and ask for a table for two, and then behind you comes somebody as a group of six. And a table for six opens up, guess what? The restaurant doesn’t put you in it because you came first, it puts the party of six at the table for six. And even though you realize they came behind you, nobody complains about it because you say, yep, that’s a better utilization of a table for six at the restaurant. That’s the kind of mindset that needs to be applied here.
Question number four, how do we know when it’s time to add chairs? Right. So how do you know when it’s time to add chairs? So if you recall I walked through the four levers you pull to get the most out of the existing chairs, one is fill the trapezoid as best as you can, two is ramp-up faster if you can, three is ramp-down more abruptly towards the end of the day if you can, and four is add some hours, then add some chairs. At some point, yes, everybody does need to add chairs. And you need to plan it with enough lead time, knowing that it takes time to get expansion built and commissioned and operational et cetera. So if you assumed it would take you six months to get that going, you need to be forecasting volume, and what you can do well enough in advance that it happens without a hiccup. Does your system communicate with any of the well-known HRs? We are completely agnostic to the HR. So our deployments span Epic, Cerner, Mosaic, [INAUDIBLE],, ARIA, all of them. And so the reason we do that, is we don’t need to integrate with any EHR. We build it so that we are not dependent on the HR at all. What our customers do is they pull the data through standard chair reports from their existing HR, whatever it is, package it like a CSV file or an Excel spreadsheet, and push it to us. All the optimization and algorithms take place with us. And then we don’t want schedulers using our system. Schedulers are very well trained on using EHR scheduling system and all the other information they need about the clinical realities of that patient are on the EHR.
So what we do is we take our recommendation and put it as templates into the EHR’s current templating mechanism. Today’s templates in all the customers– on all infusion centers that are not our customers, have been built by inspection, rights? So somebody somewhere decided that at 8 o’clock we’d start four infusions, at 8:30 we would start four more, at 9 o’clock we’d start six more. And so that template has been built with just this kind of simple minded mathematical inspection. We’re replacing that logic with the deep optimization logic that starts to say, at 7:10 you should start one 1 hour appointment and two 3 hour appointments. At 7:20 you should do one 5 hour appointment, and so on. And so we replace the simple math with sophisticated math, but otherwise co-exist with any HR, regardless. Pause for a moment. Any more questions coming up? Not seeing any come across.
MODERATOR: No I think that’s it. If you have any questions post webinar, you can still type them in and we will receive them post webinar. So if anything pops into your head, please let us know. And again, you can contact us directly using the text number at the top of your console, and also the email address that’s at the top of your console. So special thanks to Mohan for presenting today’s webinar. Friendly reminder, a recording of this webinar with the slides should appear in your inbox in about 24 hours. And thanks again to all of you for joining us today.